ADA: An Application-Conscious Data Acquirer for Visual Molecular Dynamics

Abstract

Visual molecular dynamics (VMD) has been widely used by numerous molecular dynamics (MD) applications to animate and analyze the trajectory of an MD simulation. One challenge faced by domain scientists, however, is how to filter out inactive data (i.e., data irrelevant to the subject) from the enormous output of an MD simulation. To solve it, we propose ADA (application-conscious data acquirer), a light-weight file system middleware that can perform an application-conscious data pre-processing. It provides host CPUs with only the data needed instead of an entire raw dataset. Next, we implement an ADA prototype, which is then integrated into three computing platforms: an SSD server, a nine-node OrangeFS storage cluster, and a fat-node server with 1 TB memory. Further, we evaluate ADA by running a computational biology application on the three platforms. Our experimental results show that compared to a traditional file system an ADA-assisted file system improves data processing turnaround time by up to 13.4x and reduces memory usage for data rendering by up to 2.5x. Besides, ADA allows the 1TB memory server to render more than 2x the VMD graphs while cutting energy consumption by 3x.